Recent years have brought great change in the telecommunication environment: Separate voice, video and data networks are being replaced by "broadband integrated service digital network" (B-ISDN) networks capable of supporting heterogeneous traffic. One possible protocol for implementation of B-ISDN is the Asynchronous Transfer Mode (ATM). Quality of Service (QoS) has become an important factor in the deployment of this next-generation of data networks. The continuing increase in the volume of data to be carried has boosted the need for efficient QoS administration. Although the Connection Admission Control (CAC) algorithm is not specified by the International Telecommunications Union Telecommunications (ITU-T), it is still widely used to moderate bandwidth allocation, and User Parameter Control (UPC) algorithms can ensure that contractual stipulations are met. However, if an accurate QoS monitoring technique is implemented, both the CAC and UPC mechanisms will have a firmer foundation upon which to base their decisions. QoS monitoring will allow a network operator to take an action if deterioration in the network is detected. This research focuses on the use of data interpretation to monitor the QoS of source bursty traffic based upon delay. The author has studied the monitoring process in a simulated environment of sufficient detail to produce statistically significant results. This research employs the implementation of a purpose-built simulation of an ATM network, in which the QoS experienced by different monitored sources is monitored in the presence of heterogeneous cross-traffic. The results from this simulation provide a deeper understanding of traffic interaction in broadband networks. Techniques have been deivised, tested and validated for the monitoring of both Constant Bit rate (CBR) and Variable Bit rate (VBR) traffic. The results will ultimately assist in the design of new network management strategies for ATM. Other network protocols or testing equipment will benefit from the findings of the research.

An increase of violence in public spaces has prompted the introduction of more sophisticated technology to improve the safety and security of very crowded environments. Research disciplines such as civil engineering and sociology, have studied the crowd phenomenon for years, employing human visual observation to estimate the characteristics of a crowd. Computer vision researchers have increasingly been involved in the study and development of research methods for the automatic analysis of the crowd phenomenon. Until recently, most existing methods in computer vision have been focussed on extracting a limited number of features in controlled environments, with limited clutter and numbers of people. The main goal of this thesis is to advance the state of the art in computer vision methods for use in very crowded and cluttered environments. One of the aims is to devise a method that in the near future would be of help in other disciplines such as socio-dynamics and computer animation, where models of crowded scenes are built manually on painstaking visual observation. A series of novel methods is presented here that can learn crowd dynamics automatically by extracting different crowd information from real world crowded scenes and modelling crowd dynamics using computer vision. The developed methods include an individual behaviour classifier, a scene cluttering level estimator, two people counting schemes based on colour modelling and tracking, two algorithm for measuring crowd motion by matching local descriptors, and two dynamics modelling methods - one based on statistical techniques and the other one based on a neural network. The proposed information extracting methods are able to gather both macro information, which represents the properties of the whole crowd, and micro information, which is different from individual (location) to individual (location). The statistically-based dynamics modelling models the scene implicitly. Furthermore, a method for discovering the main path of the crowded scene is developed based on it. Self-Organizing Map (SOM) is chosen in the neural network approach of modelling dynamics; the resulting SOMs are proven to be able to capture the main dynamics of the crowded scene.

Diabetes is a chronic, progressive disease that affects 1.8 million people in UK, and more than 194 million worldwide. It is currently ranked amongst the leading causes of chronic disease related death because of the occurrence of its life threatening complications. This thesis presents the design, implementation and evaluation of a mobile diabetes management and internetworking system (MDMIS). The system is based on Bluetooth and GPRS wireless communication technologies and provides an improved ubiquitous diabetes management service to both the diabetes sufferers and medical doctors. The MDMIS consists of modular suites of medical control centre, patient stations, physician stations, medical administration stations, and system maintenance stations. A patient station acquires the blood glucose measurement autonomously from a Bluetooth enabled glucose meter and transmits the data to a tailored MDMIS administration system via a GPRS wireless communications link. The medical centre of the system provides the relevant management services to both patients and physicians, such as updating user information and medication plans, side-effects reporting, analysis and alarming of blood glucose measurements, together with medical management procedures. These tasks can be accessible by patients and medics through a simple interface from various devices powered by different operating systems. The security issues of MDMIS are addressed briefly. The prototype of the system has been tested successfully. The system performance analysis of these tests is also presented in this study. The thesis also addresses the interoperability issues between mobile chronic disease management system and medical devices for universal mobile healthcare applications. An architectural framework to improve interoperability for m-health applications is presented and discussed. This architecture has been initially implemented successfully on the MDMIS system.

Over the last two decades, computer vision researchers have been working to improve the accuracy and robustness of algorithms for the context analysis of videos capturing single or multiple moving targets. However, devising algorithms that can work in uncontrolled environments with variable and unfavourable lighting conditions is still a major challenge. This thesis aims to develop robust methodologies to analyse scenes with multiple moving targets captured by a stationary camera. First, a new particle swarm optimisation algorithm is proposed to in-corporate social interaction among targets. A set of interactive swarms is employed to track multiple pedestrians in a crowd. The proposed method improves the standard particle swarm optimisation algorithm with a dynamic social model that enhances the interaction among swarms. In addition, constraints provided by temporal continuity and strength of person detections are incorporated in the tracking process. This allows the particle swarm optimisation algorithm to track multiple moving targets in a complex scene. Second, a novel method is proposed to detect global unusual events and accurately localise abnormal regions in the monitored scene. The idea is to exploit temporal coherence between video frames and use the manifold learning algorithm, in particular Laplacian Eigenmaps, to discover different crowd activities from a video. The proposed method provides an advantage of visualising and identifying different crowd events in a low dimensional space and detect abnormality. Then, this method is further extended to detect localised abnormality where the behaviour of an individual deviates from the rest of the crowd. In this approach. the visual contexts of multiple local patches are studied to model the regular behaviour of a crowded scene. This local probabilistic model allows to detect abnormal behaviour in both local and global context and localise the regions where abnormal behaviour occurs. The performance of the proposed algorithms is validated using standard data-sets and surveillance videos captured in uncontrolled environments.

It is expected nowadays that robots are able to work in real-life environments, possibly also sharing the same space with humans. These environments are generally considered as being cluttered and hard to train for. The work presented in this thesis focuses on developing an online and real-time biologically inspired model for teams of robots to collectively learn and memorise their visual environment in a very concise and compact manner, whilst sharing their experience to their peers (robots and possibly also humans). This work forms part of a larger project to develop a multi-robot platform capable of performing security patrol checks whilst also assisting people with physical and cognitive impairments to be used in public places such as museums and airports. The main contribution of this thesis is the development of a model which makes robots capable of handling visual information, retain information that is relevant to whatever task is at hand and eliminate superfluous information, trying to mimic human performance. This leads towards the great milestone of having a fully autonomous team of robots capable of collectively surveying, learning and sharing salient visual information of the environment even without any prior information. Solutions to endow a distributed team of robots with object detection and environment understanding capabilities are also provided. The way in which humans process, interpret and store visual information are studied and their visual processes are emulated by a team of robots. In an ideal scenario, robots are deployed in a totally unknown environment and incrementally learn and adapt to operate within that environment. Each robot is an expert of its area however, they possess enough knowledge about other areas to be able to guide users sufficiently till another more knowledgeable robot takes over. Although not limited, it is assumed that, once deployed, each robot operates in its own environment for most of its lifetime and the longer the robots remains in the area the more refined their memory will become. Robots should to be able to automatically recognize previously learnt features, such as faces and known objects, whilst also learning other new information. Salient information extracted from the incoming video streams can be used to select keyframes to be fed into a visual memory thus allowing the robot to learn new interesting areas within its environment. The cooperating robots are to successfully operate within their environment, automatically gather visual information and store it in a compact yet meaningful representation. The storage has to be dynamic, as visual information extracted by the robot team might change. Due to the initial lack of knowledge, small sets of visual memory classes need to evolve as the robots acquire visual information. Keeping memory size within limits whilst at the same time maximising the information content is one of the main factors to consider.

The expectation from a future generation cellular network is to provide multiplay applications of VoIP, video and data to a continuously growing number of cellular users. The scarcity of the available radio spectrum coupled with the unique traffic handling and Quality of Experience (QoE) requirements of the converged services poses a huge challenge to the network operators. The solution of over-provisioning the network by increasing the amount of bandwidth is not economical. Therefore, efficient partition of network resources becomes mandatory. Scheduling plays an important in determining the overall efficiency of a wireless system. This thesis focuses on quality driven scheduling for efficient resource allocation in multi-user downlink LTE systems. Video traffic contributes a major proportion of network traffic. Therefore, one of the main goals of this work is to design scheduling strategies which consider information about video traffic with the aim of improving the service quality perceived by the user. Various scheduling strategies are proposed taking into account different criteria such as packet delay and importance of a video packet. This thesis presents a novel cross-layer resource allocation architecture which reduces the need for cross-layer signaling and frequent end-to-end link probing (for video rate adaptation) required by other cross-layer approaches. Apart from the novel cross-layer architecture, the thesis applies the concepts of game theory and fuzzy logic frameworks in radio resource management and proposes a composite scheduling rule which considers the service needs of different traffic types such as video, VoIP and data. Results show that the proposed scheduling schemes lead to an efficient partition of radio resources while achieving a significant improvement in the perceived quality as compared to state-of-the-art scheduling rules.

In this thesis we propose novel methods for 3D interaction on 3D environments. The evaluation of these methods was performed based on three interaction environments: 3D interaction using portable multi-touch devices, 3D hand gesture data manipulation using 3D database representation and 3D multi-threaded programing using hand gesture interaction. The three experiments provided qualitative and quantitative information to evaluate the features of the presented interfaces. The first experiment, based on the use on the use of portable multi-touch devices, aimed to evaluate the use of 3D movements to interact under a 3D environment. Also, the possibility of generate collaborative interaction under 3D interfacing (simulating a 3D multi-touch table top environment) was evaluated. The second experiment consisted on 3D touchless data manipulation, removing the intermediate device (portable multi-touch) and providing hand gesture data interaction using the Kinect device. Furthermore, this evaluation was conducted over a 3D cube database model, based on the concepts of multidimensional databases and graphic databases. The third experiment intended to evaluate the possibility of software generation using a 3D interaction environment, following a similar model of interaction from the second experiment, but providing a better two handed interaction. The environment aimed multi-threaded programing under a 3D interface. The three experiments provided valuable data about users’ interaction and preference, which were tested with users of different ages and levels of knowledge. The research process and results are summarized in this research work.

The aim of this thesis is to investigate techniques for accurate segmentation of the hippocampus in order to measure the degree of atrophy associated with diseases such as Alzheimer’s, temporal lobe epilepsy, long-lasting traumatic stress and schizophrenia. To this end, specific algorithms and methodologies are developed to segment the hippocampus from structural magnetic resonance (MR) images, in combination with pre- and post-processing operations to improve robustness and accuracy. Segmentation efficiency is boosted by pre-processing the input image with a bias correcting spatial fuzzy c-means algorithm and a nonlocal mean filter to smooth the MRI dataset whilst preserving edges. A 3D level set method is used to segment the left and right hippocampi simultaneously. The thesis investigates the problem of initialisation of the level set algorithm, which must cope with some challenging characteristics of the hippocampus, such as the small size, wide range of internal intensities, narrow width, and shape variation. Due to intensity inhomogeneity, using a single seed region inside the hippocampus is prone to failure. Hence, alternative initialisation strategies are explored, such as using multiple initialisations in different sections (such as the head, body and tail) of the hippocampus and ‘tailored’ initialisation based on superquadrics. Accurate quantification of a segmented hippocampus can provide essential details for diagnosis, treatment planning, and follow-up comparisons. Hence, a post-processing approach to quantify the partial volume effect (PVE) for correction of the hippocampal volume is assessed. The method enables estimation of the PVE in order to generate more accurate measurements of the hippocampal volume. The results of segmentation are evaluated on two public MRI datasets that include annotated ground-truth to identify the hippocampus. Experimental results indicate that using a single initialisation results in an average correct segmentation of only 39%, though the performance rises to 85% when using the multiple initialisations approach. These results are shown to exceed the performance achieved by other researchers for these datasets. The analyses of corrected volumes of the several publicly available datasets are used to quantify the asymmetry in the size of the left and right hippocampi. The measure of asymmetry is applied to a set of normal scans and ones from epileptic patients. The average asymmetry values were 7% and 12% respectively, indicating asymmetry may be a useful index for diagnosis of diseases associated with the differential shrinkage of the hippocampus.

Crowd analysis from video footage is an active research topic in the field of computer vision. Crowds can be analaysed using different approaches, depending on their characteristics. Furthermore, analysis can be performed from footage obtained through different sources. Fixed CCTV cameras can be used, as well as cameras mounted on moving vehicles. To begin, a literature review is provided, where research works in the the fields of crowd analysis, as well as object and people tracking, occlusion handling, multi-view and sensor fusion, and multi-target tracking are analyses and compared, and their advantages and limitations highlighted. Following that, the three contributions of this thesis are presented: in a first study, crowds will be classified based on various cues (i.e. density, entropy), so that the best approaches to further analyse behaviour can be selected; then, some of the challenges of individual target tracking from aerial video footage will be tackled; finally, a study on the analysis of groups of people from multiple cameras is proposed. The analysis entails the movements of people and objects in the scene. The idea is to track as many people as possible within the crowd, and to be able to obtain knowledge from their movements, as a group, and to classify different types of scenes. An additional contribution of this thesis, are two novel datasets: on the one hand, a first set to test the proposed aerial video analysis methods; on the other, a second to validate the third study, that is, with groups of people recorded from multiple overlapping cameras performing different actions.

Tracking pedestrians in surveillance videos is an important task, not only in itself but also as a component of pedestrian counting, activity and event recognition, and scene understanding in general. Robust tracking in crowded environments remains a major challenge, mainly due to the occlusions and interactions between pedestrians. Methods to detect humans in a single frame are becoming increasingly accurate. Therefore, the majority of multi-target tracking algorithms in crowds follow a tracking-by-detection approach, along with models of individual and group behaviour, and various types of features to re-identify any given pedestrian (and discriminate them from the remainder). The aim is, given a Closed Circuit TeleVision (CCTV) camera view (moving or static) of a crowded scene, to produce tracks that indicate which pedestrians are entering and leaving the scene to be used in further applications (e.g. a multi-camera tracking scenario). Therefore, this output should be accurate in terms of position, have few false alarms and identity changes (i.e. tracks have not to be fragmented nor switch identity). Consequently, the presented algorithm concentrates on two important characteristics. Firstly, production of a real-time or near real-time output to be practically usable for further applications without penalising the final system. Secondly, management of occlusions which is the main challenge in crowds. The methodology presented, based on a tracking-by-detection approach, proposes an advance over those two aspects through a hierarchical framework to solve short and long occlusions with two novel methods. First, at a fine temporal scale, kinematic features and appearance features based on non-occluded parts are combined to generate short and reliable 'tracklets'. More specifically, this part uses an occlusion map which attributes a local measurement (by searching over the non-occluded parts) to a target without a global measurement (i.e. a measurement generated by the global detector), and demonstrates better results in terms of tracklet length without generating more false alarms or identity changes. Over a longer scale, these tracklets are associated with each other to build up longer tracks for each pedestrian in the scene. This tracklet data association is based on a novel approach that uses dynamic time warping to locate and measure the possible similarities of appearances between tracklets, by varying the time step and phase of the frame-based visual feature. The method, which does not require any target initialisations or camera calibrations, shows significant improvements in terms of false alarms and identity changes, the latter being a critical point for evaluating tracking algorithms. The evaluation framework, based on different metrics introduced in the literature, consists of a set of new track-based metrics (in contrast to frame-based) which enables failure parts of a tracker to be identified and algorithms to be compared as a single value. Finally, advantages of the dual method proposed to solve long and short occlusions are to reduce simultaneously the problem of track fragmentation and identity switches, and to make it naturally extensible to a multi-camera scenario. Results are presented as a tag and track system over a network of moving and static cameras. In addition to public datasets for multi-target tracking in crowds (e.g. Oxford Town Centre (OTC) dataset) where the new methodology introduced (i.e. building tracklets based on non-occluded pedestrian parts plus re-identification with dynamic time warping) shows significant improvements. Two new datasets are introduced to test the robustness of the algorithm proposed in more challenging scenarios. Firstly, a CCTV shopping view centre is used to demonstrate the effectiveness of the algorithm in a more crowded scenario. Secondly, a dataset with a network of CCTV Pan Tilt Zoom (PTZ) cameras tracking a single pedestrian, demonstrates the capability of the algorithm to handle a very difficult scenario (abrupt motion and non-overlapping camera views) and therefore its applicability as a component of a multitarget tracker in a network of static and PTZ cameras. The thesis concludes with a critical analysis of the work and presents future research opportunities (notably the use of this framework in a non-overlapping network of static and PTZ cameras).